Author: leonid
Date: 18:53:11 11/19/00
Go up one level in this thread
On November 19, 2000 at 13:09:06, Uri Blass wrote: >On November 19, 2000 at 12:44:38, leonid wrote: > >>On November 19, 2000 at 01:28:42, Uri Blass wrote: >> >>>On November 18, 2000 at 17:08:01, leonid wrote: >>> >>>>On November 18, 2000 at 15:48:12, Timothy J. Frohlick wrote: >>>> >>>>>On November 18, 2000 at 13:39:39, Jorge Pichard wrote: >>>>> >>>>>>Does anybody know which programs use which method and which aproach is yielding >>>>>>a better result? >>>>> >>>>>All of todays chess programs use various forms of selective search. Our >>>>>processors could not even begin to do full-width and complete 14 ply searches. >>>>>Brute force machines would need processors and memory millions of times as much >>>>>as even the Pentium IV 1.5 Ghz machines that just arrived today. >>>> >>>> >>>>As far as I know heavy position could be taken by brute force in few seconds on >>>>1.5 Ghz only 8 plys deep. Heavy position - average 35 nodes in each ply. >>>>8 ply search is too little. At least, actual computer must be able to reach some >>>>12 plys to make brute force search practical. We are still not there. >>> >>>I believe that brute force is going to lose against selctive search of tiger and >>>it is not important if the number of plies of the brute force program is 8 or >>>12. >>> >>>You can do experiment by giving your brute force program to search 12 plies and >>>giving the selective search programs to search for the same time. >>> >>>12 plies may take some hours for the brute force program so the game will be >>>correspondence time control game. >> >>So, probably you want to say the same thing as me, brute force search is too >>slow right now. But one thing is clear, if one program will search 12 plys by >>brute force and the next by selective, selective have better chance to lose. And >>if both programs do this search in time of 1 second? Brute force just in one >>second and selective in a split of one second? What is the difference? But for >>now we a not there. >> >> >>>> >>>>>Full width or "brute force" is not a smart way to play chess. Kramnik and >>>>>Kasparov are not brute force players. >>>> >>>>Brute force is not affordable way of playing. This is only reason why it look >>>>like very poor. This impression could become different only in some 30 years >>>>from now. We are still playing with computers of Stone Age. >>> >>>I do not expect brute force to become the right way in the future. >>>Brute force can solve the game but my guess is that being 1000 times faster or >>>100000 times faster is not enough for getting better results relative to >>>selective search. >> >>Disagree, Uri! Brute force search will be good enough not when it will be better >>that selective search (or any other kind of search) but when it will start >>beating all human around. > >I disagree. Probably you are disagree because our conversation is too wide. If we will go to see some practical details it could be that we speak the same. If talking goes about game where each move must take no more that few seconds, it will be more clear. I like mainly this kind of games. So, how deep usually in the game mate existe and how far excellent brain can see them? After what I could find (not personal impression but long practical research) mate is usually no more deep that 2 or 4 moves deep (4 - 8 plys). Very seldom 5 moves. So when I looked what kind of search I must put into mate checking I looked how much time 5 moves brute force can take. It last too long. This is the reason why 7 moves selective search was used. But when 5 moves brute force will be able to look each position in one second, selective search will be no more needed. Human, even the best one, can hardly see mate in few seconds 5 moves ahead. Selective search of 30 plys deep is nothing but overkill. The same should be true for all other part of chess game beyond mate search. Only there depth of seeing in few seconds for human is probably different. >There are people who are interested in comp-comp games and beating all the >humans is not good enough and the target is to beat also all the computers. I had in my mind human more as participant and not as spectator. But for computers as players, I agree. > And when we forget about openings and the end of the >>game (there data will be needed) human can see only up to the fixed depth in the >>game at limited time. Once computer will start seeing over this limit, all other >>extra strength will be added for sheer vanity. Then brute force search will >>become real. > >I again disagree. >Humans can see more than 12 plies forward in some selective lines. Given short time, not that much. And in general, in the game where each move must take only few seconds, once human only sometime can seen beyond 12 plys but almost all the time can't do so, it is a big looser. >Computers are better than humans in short tactics because they see everything at >short depth but the best humans can be better in long tactics and there are >cases that you need to search 30 plies forward or more than it in the relevant >lines. So, it is quite possible to make program that will do its search both ways. One for "normal game", where you don't want to play for ever and where one move take only few seconds. There brute force is everything that program will need. Second one for slow, or deadly slow (professional) playing. There selective search should be used to find needed move. But even for second kind of playing everything in program's algorith could be simplified. Brute force of engin will make this possible. Leonid >Uri
This page took 0 seconds to execute
Last modified: Thu, 15 Apr 21 08:11:13 -0700
Current Computer Chess Club Forums at Talkchess. This site by Sean Mintz.